The $400 Million Machine That Could Decide the AI Chip Race

Nvidia can design an AI accelerator. A chip manufacturer still has to turn that design into physical silicon.

That requires many kinds of equipment, but one category is unusually hard to reproduce: the machines that print the smallest patterns onto a wafer. The most advanced of these use extreme ultraviolet light, or EUV. ASML is the only commercial supplier of EUV lithography systems today (see Sources, [1], [3]).

Its next-generation High NA systems are reported to cost around $400 million each. That is roughly twice the reported price of a conventional EUV system, and it raises a practical question: why would a chipmaker commit to a machine this expensive when it can keep improving other parts of the manufacturing process? (See Sources, [1].)

The answer is about resolution, production economics, and the difficulty of finding an alternative at the leading edge.

The design is not the chip

A chip design describes the circuits that should exist. Manufacturing must reproduce those circuits across silicon wafers, layer by layer, with extraordinary precision.

Lithography projects a patterned image onto a light-sensitive coating on the wafer. Other steps then develop, etch, deposit, and measure the structures. The sequence repeats many times before a functioning chip emerges.

Not every layer requires the most advanced equipment. ASML says EUV is used for the most intricate layers of leading-edge chips, while deep ultraviolet, or DUV, systems print many of the other layers (see Sources, [3]).

That distinction matters. ASML is not the only maker of lithography machines in general. Its particularly strong position is in commercial EUV systems, the specialized tools needed for some of the smallest patterns in advanced manufacturing.

A processor designer can compete on architecture, software, and packaging. Yet its manufacturing partner must also have the equipment, process knowledge, and capacity to print the design at a useful yield.

How EUV turns tin into light

Older DUV lithography uses light with a wavelength of 193 nanometers. EUV uses a much shorter wavelength: 13.5 nanometers. The shorter wavelength helps print finer patterns, although actual resolution also depends on the optics and manufacturing process (see Sources, [3]).

There is no ordinary EUV lamp inside the system.

ASML describes firing two laser pulses at a fast-moving droplet of tin. The resulting plasma emits EUV light. Its EUV systems repeat this process up to 50,000 times a second. Mirrors guide the light through the pattern on a reticle and toward the wafer (see Sources, [3]).

This is difficult because EUV light is absorbed by air and many ordinary materials. The light path must remain in a high vacuum, and conventional refractive lenses cannot do the job. ASML uses multilayer reflective optics instead (see Sources, [3]).

The machine is therefore more than a powerful light source. It is a coordinated system of lasers, vacuum equipment, mirrors, moving stages, control software, and precision manufacturing. A weakness in any one part can limit what the full tool can deliver.

Why a smaller pattern can be worth a larger machine

The established NXE generation uses a numerical aperture, or NA, of 0.33. High NA raises it to 0.55. NA is a measure related to how much of the light the optical system can collect and use to form the image. In this context, the change helps improve resolution (see Sources, [3], [4]).

ASML says its EXE:5000 High NA system can print features 1.7 times smaller in a single exposure than its NXE systems and can enable transistor densities 2.9 times higher. Those are statements about the system's imaging capability and potential chip designs, not a promise that every manufactured chip will automatically contain 2.9 times as many working transistors (see Sources, [4]).

A finer single exposure may also replace some more complicated patterning steps. In principle, fewer steps can reduce time, defects, and cost. Whether those savings justify a particular machine purchase depends on each chipmaker's designs, yields, production volume, and process choices.

That is the economic calculation behind the reported $400 million price. The customer is buying a path to future manufacturing capability, not simply a larger version of a current machine.

A small number of machines, a large business

ASML's scale looks different from the scale of a consumer electronics company.

In its 2025 annual report, ASML recorded €32.7 billion in total net sales, €4.7 billion in research and development, and 48 EUV lithography systems among its recognized system sales (see Sources, [2]).

The 48 figure covers EUV systems in the annual report; it should not be read as 48 High NA systems. It illustrates the economics of a company selling a limited number of extraordinarily complex machines, alongside other systems, software, and services.

A chipmaker cannot add leading-edge capacity just by ordering more GPUs. It needs fab space, people, materials, process development, and a working lithography fleet. Each tool must be built, delivered, installed, and qualified for the intended manufacturing process.

The equipment business therefore sits upstream of the AI chip business. Demand for accelerators can turn into demand for foundry capacity, which in turn can become demand for lithography systems years before the chips reach a data center.

What $400 million does and does not guarantee

Reuters reported in September 2026 that established EUV tools were around $200 million and the new High NA generation around $400 million. It also reported that Intel, TSMC, Samsung, and SK Hynix had varying plans to use High NA, with TSMC targeting adoption around 2030 (see Sources, [1]).

Those are reported prices and company plans, not a universal catalog price or a guarantee that every customer will deploy on the same schedule.

The reasons to wait are real. A newer tool needs process development, compatible materials, masks, integration with other equipment, and enough yield to earn back its cost. A manufacturer may instead extend current EUV processes, use multiple patterning, or improve packaging and chiplet designs for a particular product.

That makes High NA a strategic option, rather than a simple switch that the entire industry flips at once. Its value depends on the chip and the economics of producing it.

The earlier Engines & Margins article The Hidden Infrastructure Companies Behind the AI Boom follows the companies supplying power, cooling, and connectivity after a chip is made. Lithography is further upstream: it helps determine which designs can become affordable physical chips at scale.

Where the bottleneck sits

ASML's position is strongest in EUV. Reuters reported that the company has no commercial EUV rival, while other suppliers make DUV tools. It also reported strong commitments for the High NA generation (see Sources, [1]).

That creates a concentration risk for the wider industry. If advanced-chip demand rises faster than EUV systems can be produced and installed, chipmakers cannot instantly substitute an equivalent tool from another commercial supplier.

But the bottleneck is not controlled by one machine alone. The chipmaker still needs a successful process, sufficient yield, packaging capacity, memory, and the power and facilities to run the finished processors. ASML is one critical part of a much larger production chain.

This is why the title says the machine could decide part of the AI chip race. Better lithography may help manufacturers make smaller, more efficient circuits economically. The outcome also depends on the designs, materials, process engineering, and competing ways to improve performance.

The AI boom may be discussed as a contest between chip designers. At the factory, it is also a contest to turn increasingly ambitious designs into working silicon. ASML's EUV systems are among the hardest tools in that chain to replace.